Multivariate Regression With Dependence Structures: Evaluating Associations Between Plasma Metabolomics and Alcohol...

Area of Science:

• Biostatistics
• Bioinformatics
• Systems Biology

Background:

• High-dimensional omics data display complex network structures (modularity, small-worldness, scale-free).
• Integrating these network properties into multivariate regression is challenging due to estimation difficulties.
• Existing covariance methods may fail to preserve network structures and are computationally intensive.

Purpose of the Study:

• To develop a novel multivariate regression model that incorporates interconnected community structures from omics data.
• To create efficient estimation algorithms for regression and dependence parameters.
• To ensure theoretical robustness and accurate hypothesis testing for omics network analysis.

Main Methods:

• Proposed a multivariate regression model with interconnected community structure.
• Developed closed-form regression and likelihood-based dependence estimators.
• Established asymptotic properties for estimator robustness and hypothesis testing.
• Conducted extensive simulations for accuracy and sensitivity benchmarking.

Main Results:

• The novel method demonstrated enhanced accuracy and sensitivity compared to existing models in simulations.
• Applied to metabolomic data (249 biomarkers) from 3984 participants.
• Light alcohol consumption showed positive associations with high-density lipoprotein cholesterol (HDL), HDL particles, and Apolipoproteins A1.

Conclusions:

• The proposed model effectively integrates omics network structures into regression analysis.
• The findings suggest light alcohol intake is linked to improved cardiovascular health indicators.
• This approach offers a robust framework for analyzing complex biological networks and their associations.